Search papers, labs, and topics across Lattice.
This paper introduces a multimodal-prior-guided importance sampling technique for hierarchical 3D Gaussian Splatting (3DGS) to improve sparse-view novel view synthesis. The method fuses photometric rendering residuals, semantic priors, and geometric priors to estimate local recoverability, guiding the injection of fine Gaussians in a coarse-to-fine manner. By focusing refinement on geometrically critical regions supported by multimodal evidence, the approach reduces overfitting and noise, leading to state-of-the-art reconstruction results.
By intelligently fusing semantic and geometric priors with rendering residuals, this method achieves state-of-the-art novel view synthesis from sparse inputs, outperforming methods relying solely on photometric error.
We present multimodal-prior-guided importance sampling as the central mechanism for hierarchical 3D Gaussian Splatting (3DGS) in sparse-view novel view synthesis. Our sampler fuses complementary cues { -- } photometric rendering residuals, semantic priors, and geometric priors { -- } to produce a robust, local recoverability estimate that directly drives where to inject fine Gaussians. Built around this sampling core, our framework comprises (1) a coarse-to-fine Gaussian representation that encodes global shape with a stable coarse layer and selectively adds fine primitives where the multimodal metric indicates recoverable detail; and (2) a geometric-aware sampling and retention policy that concentrates refinement on geometrically critical and complex regions while protecting newly added primitives in underconstrained areas from premature pruning. By prioritizing regions supported by consistent multimodal evidence rather than raw residuals alone, our method alleviates overfitting texture-induced errors and suppresses noise from pose/appearance inconsistencies. Experiments on diverse sparse-view benchmarks demonstrate state-of-the-art reconstructions, with up to +0.3 dB PSNR on DTU.
